fix(subtitle): unify create() return value and check result in caller

[yu3394]

• Add return type annotation -> str to subtitle.create()
• Return empty string instead of None when model loading fails, consistent with import failure
• Return subtitle_file path on success so callers can verify generation
• Check create() return value in task.py before calling correct(), skip correction and return early on failure

[harry0703]

Thank you for fixing the inconsistent Whisper subtitle result contract. The production change is small and the caller-side guard is the right direction, but the PR is not yet merge-ready because the repository tests were not updated with the new contract.

My local run reports 2 failures, 53 passes, and 3 skips:

• test_create_returns_none_when_whisper_model_cannot_load still expects None instead of the new empty-string failure value.
• test_generate_subtitle_uses_whisper_for_custom_audio_without_sub_maker uses a successful fake create() that writes the SRT but returns None; the new caller therefore treats it as failure and skips correction.

Please update those tests, make the successful fake return subtitle_file, and add explicit coverage that an empty create() result skips correct() and returns an empty subtitle path. It would also be useful to assert that a successful real/fake create() returns the generated path. Ruff already passes.

Thanks again—once the tests reflect the new contract, this should be a low-risk reliability improvement.